Hi all,

I have a doubt about the bellow scenario, where the ISP1 use eBGP

sessions to its peers and is a BGP Transit AS.

NSP 1 ------------------ ISP 1 Router2 ----------- NSP 2

> >

> >

> >

> annunce /21 |

> >

Customer1 --------------- ISP 1 Router1

announce /20

The "Customer1" is client on both ISPs (ISP1 and NSP1) and have an /20

IP prefix. To NSP1, it announce two /21 prefixes. To ISP1, it announce

a /20 prefix. If traffic comes from NSP 2 (connected only to ISP 1) to

Customer1, the ISP 1 Routers try to send data over NSP 1, ignoring the

Custormer1->ISP1 link.

To solve this question, an solution that I found is filter Customer1

prefixes in BGP session between NSP1 and ISP1 Router2. But this don't

appear scalable...

Is this solution right ? What is the better solution for this

scenario? How large ISPs solve this kind of problem?

Thanks,

Rafael

The more specific /21's are winning over the /20, so they will always be preferred by default. If you want to change that, you could announce the /20 to NSP1, or announce the /21's to ISP1.

Mike

Hi all,

I have a doubt about the bellow scenario, where the ISP1 use eBGP

sessions to its peers and is a BGP Transit AS.

NSP 1 ------------------ ISP 1 Router2 ----------- NSP 2

> >

> annunce /21 |

> >

Customer1 --------------- ISP 1 Router1

announce /20

The "Customer1" is client on both ISPs (ISP1 and NSP1) and have an /20

IP prefix. To NSP1, it announce two /21 prefixes. To ISP1, it announce

a /20 prefix. If traffic comes from NSP 2 (connected only to ISP 1) to

Customer1, the ISP 1 Routers try to send data over NSP 1, ignoring the

Custormer1->ISP1 link.

To solve this question, an solution that I found is filter Customer1

prefixes in BGP session between NSP1 and ISP1 Router2. But this don't

appear scalable...

longest match wins...

if you're customer 1

deaggregate the avertisement to isp 1 or re-aggregate the advertisement to nsp 1. either will achieve the same end.

if you're isp1 consider what customer 1 was trying to achieve by doing this. e.g. they're traffic engineering (or they are clueless) and theirfore have a vested interest in the current path.

Dear Rafael,

Is this solution right ? What is the better solution for this

scenario? How large ISPs solve this kind of problem?

communitie(filters) help to scale.

for example lambdanet communities:

remarks: Prepend communities to modify announcements to peers

remarks:

remarks: 13237:3811n announcements to AS12322 (Free)

remarks: 13237:3812n announcements to AS3356 (Level3)

remarks: 13237:3813n announcements to AS8220 (COLT)

remarks: 13237:3814n announcements to AS286 (KPN Eurorings)

remarks: 13237:3815n announcements to AS3303 (Swisscom)

remarks: 13237:3818n announcements to AS9121 (TurkTelekom)

remarks: 13237:3824n announcements to AS2914 (Verio)

remarks: 13237:3825n announcements to AS4766 (Korea Telecom)

remarks: 13237:3826n announcements to AS3491 (BtN)

remarks: 13237:3828n announcements to AS8928 (Interoute)

remarks: 13237:3830n announcements to AS1257 (Swipnet)

remarks: 13237:3831n announcements to AS3292 (TeleDanmark)

remarks: 13237:3832n announcements to AS3209 (Arcor)

remarks: 13237:3833n announcements to AS3320 (DTAG)

remarks: 13237:3835n announcements to AS6805 (Telefonica DE)

remarks: 13237:3836n announcements to AS8447 (Telekom AT)

remarks: 13237:3837n announcements to AS8881 (Versatel DE)

remarks: 13237:3838n announcements to AS13184 (Hansenet)

remarks: 13237:3855n announcements to AS6830 (Chello)

remarks: 13237:3860n announcements to AS3257 (Tiscali Int.)

remarks: 13237:3865n announcements to AS702 (MCI EU)

remarks: 13237:3866n announcements to AS3549 (Global Crossing)

remarks: 13237:3869n announcements to AS6453 (Tata/Teleglobe)

remarks: 13237:3870n announcements to AS20676 (QSC)

remarks: 13237:3876n announcements to AS2856 (BT UK)

remarks: 13237:3877n announcements to AS2119 (Telenor)

remarks: 13237:3891n announcements to AS1299 (TeliaSonera)

remarks: 13237:3892n announcements to AS6461 (Abovenet)

remarks:

remarks: with n = 0,1,2,3 meaning

remarks: n = 0 do not announce to peer

remarks: n = 1 prepend "AS13237"

remarks: n = 2 prepend "AS13237 AS13237"

remarks: n = 3 prepend "AS13237 AS13237 AS13237"

Kind regards,

Ingo Flaschberger